/MHN

ICCV2019

Primary LanguagePython

Code for ICCV 2019 Paper "Mixed High-Order Attention Network for Person Re-Identification"

This code is developed based on pytorch framework and the baseline code.

Updates

  • Aug 16, 2019

    • The codes of training and testing for our ICCV19 paper are released.
    • We have cleared up and tested the codes on Market, Duke datasets, the expected retrieval performances are as follows:
    Market R@1 R@5 R@10 mAP Reference
    IDE+ERA 89.9% 96.4% 97.6% 75.6% train_ide.py
    IDE+MHN6 93.1% 97.7% 98.7% 83.2% train_ide.py
    PCB+ERA 91.7% 97.4% 98.3% 76.4% train_smallPCB
    PCB+MHN4 94.3% 98.0% 98.8% 83.9% train_smallPCB
    PCB+MHN6 94.8% 98.3% 98.9% 85.2% train_smallPCB_multiGPU.py
    Duke R@1 R@5 R@10 mAP Reference
    IDE+ERA 82.7% 91.8% 94.1% 68.1% train_ide.py
    IDE+MHN6 87.8% 94.2% 95.8% 74.6% train_ide.py
    PCB+ERA 82.9% 91.7% 93.8% 67.7% train_smallPCB
    PCB+MHN4 88.5% 94.5% 96.1% 76.9% train_smallPCB
    PCB+MHN6 89.5% 94.7% 96.1% 77.5% train_smallPCB_multiGPU.py

Files

  • train_ide.py test_ide.py

    • files for training and testing on IDE framework

  • train_smallPCB.py test_smallPCB.py

    • files for training and testing on PCB framework, when using MHN, the maximized order is limited to 4 due to the GPU memory.

  • train_smallPCB_multiGPU.py test_smallPCB.py

    • files for training, if you want to test MHN6, please use this file for training with multi gpus. The testing file is also test_smallPCB.py

  • auto_test.sh

    • auto-testing code.

Prerequisites

  • Pytorch(0.4.0+)
  • python3.6
  • 2GPUs, each > 11G

Train_Model

  1. Clone our code.
  2. Download the training images {google drive, baidu}, including Market1501, DukeMTMC, CUHK03-NP.
  3. Go into the MHN/ dir and mkdir datasets/, then unzip the downloaded datasets.zip to datasets/
  4. Run prepare.py to preprocess the datasets.
  5. Then you can try our methods
IDE+ERA
python3 train_ide.py --gpu_ids 0 --name ide --data_dir datasets/Market/datasets/pytorch/ --train_all --batchsize 32 --erasing_p 0.4 --balance_sampler
IDE+MHN6
python3 train_ide.py --gpu_ids 0 --name ide_mhn6 --data_dir datasets/Market/datasets/pytorch/ --train_all --batchsize 32 --erasing_p 0.4 --balance_sampler --alpha 1.4 --parts 6 --mhn
PCB+ERA
python3 train_smallPCB.py --gpu_ids 0 --name pcb --data_dir datasets/Market/datasets/pytorch/ --train_all --batchsize 32 --erasing_p 0.4 --balance_sampler
PCB+MHN4
python3 train_smallPCB.py --gpu_ids 0 --name pcb_mhn4 --data_dir datasets/Market/datasets/pytorch/ --train_all --batchsize 32 --erasing_p 0.4 --balance_sampler --alpha 2 --parts 4 --mhn
PCB+MHN6
python3 train_smallPCB_multiGPU.py --gpu_ids 0,1 --name pcb_mhn6 --data_dir datasets/Market/datasets/pytorch/ --train_all --batchsize 32 --erasing_p 0.4 --balance_sampler --alpha 2 --parts 6 --mhn

the trained models are stored in folder "model/($name)".

Evaluation

We provide the auto-testing code in auto_test.sh, you can replace the corresponding code for testing. For example,

For IDE+ERA
python3 test_ide.py --gpu_ids $gpu_ids --name ide --test_dir datasets/Market/datasets/pytorch/ --batchsize 32 --which_epoch $i
For IDE+MHN6
python3 test_ide.py --gpu_ids $gpu_ids --name ide_mhn6 --test_dir datasets/Market/datasets/pytorch/ --batchsize 20 --which_epoch $i --mhn --parts 6
For PCB+ERA
python3 test_smallPCB.py --gpu_ids $gpu_ids --name pcb --test_dir datasets/Market/datasets/pytorch/ --batchsize 32 --which_epoch $i
For PCB+MHN4
python3 test_smallPCB.py --gpu_ids $gpu_ids --name pcb_mhn4 --test_dir datasets/Market/datasets/pytorch/ --batchsize 15 --which_epoch $i --mhn --parts 4
For PCB+MHN6
python3 test_smallPCB.py --gpu_ids $gpu_ids --name pcb_mhn6 --test_dir datasets/Market/datasets/pytorch/ --batchsize 10 --which_epoch $i --mhn --parts 6

Contact

Citation

You are encouraged to cite the following papers if this work helps your research.

@inproceedings{chen2019mixed,
  title={Mixed High-Order Attention Network for Person Re-Identification},
  author={Chen, Binghui and Deng, Weihong and Hu, Jiani},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision (ICCV)},
  year={2019},
}
@InProceedings{chen2019energy,
author = {Chen, Binghui and Deng, Weihong},
title = {Energy Confused Adversarial Metric Learning for Zero-Shot Image Retrieval and Clustering},
booktitle = {AAAI Conference on Artificial Intelligence},
year = {2019}
}

License

Copyright (c) Binghui Chen

All rights reserved.

MIT License

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.